Performance of Portforio Selection Method on the Concept of Neural Network
Hidenori Naoe · IEEJ Transactions on Electronics Information and Systems · 1999
The present paper proposes a portfolio selection method on the concept of neural network, which selects a few stocks with low risk and high returns among the many. It was difficult to solve this problem even with a large computer, because the number of the feasible combination of the stocks is enormous. Applying the concept of neural network, quasi-minimum points of the energy function are searched and the quasi-optimal stocks are obtained. The measure of the performance is the deviation from the mean. Then, the performance of the solution by the simulated annealing method using the Hopfield network is compared with that by the simulated annealing method with the Lagrangian multiplier, by the greedy method, by the all searching method and by the random method. It is concluded that the performance of these several methods is almost the same. The risk and the return of the portfolios have the trade-off relation. Consequently, it is possible to apply the simulated annealing method using the Hopfield network in order to select portfolios